Shift/esd

There is no general answer to this question. Too few cycles lead to incomplete convergence and four or six seems to be the minimum even in very well-behaved refinements. Too many cycles will not hurt the refinement but can waste computing time. I usually use 10 cycles less if a refinement converges very well, and more when needed. You can tell that a refinement converges when the values for Mean shift/esd and Maximum become very small (ideally 0.000, but 0.01 or even 0.1 is acceptable in early stages of the refinement). 

Fig. 20. Calculated dependence of the value of relative shift [Esd (L) - Esa (1)]/AEsa (oo) of Ge 3d level for cluster (continuous curve) and continuous layer (bottom dashed curve) on characteristic L parameter radius for a cluster and thickness for a layer. Transition to continuous monociystal layer during recrystaUization is shown with an arrow specimen temperature is shown over the experimental points.

---